# US 12,764,093 B2 · Providing explainable artificial intelligence using distributed ledger technology

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,764,093 B2
- Document ID: US12764093B2
- Status: Granted
- Kind: B2
- Application number: 19/083,983
- Filed: 2025-03-19
- Published: 2025-07-03
- Granted: 2026-10-13
- Earliest claimed: 2023-09-20
- Assignees: Bank of America Corporation
- Applicants: Bank of America Corporation
- Inventors: Maharaj Mukherjee
- Web page: https://patentgrep.com/patent/US12764093B2

## CPC

- H04L 63/0209: {Architectural arrangements, e.g. perimeter networks or demilitarized zones}
- G01D 21/02: Measuring two or more variables by means not covered by a single other subclass
- G06F 40/30: Semantic analysis
- G06N 20/00: Machine learning
- H04L 9/50: {using hash chains, e.g. blockchains or hash trees}
- H04L 41/16: using machine learning or artificial intelligence
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G01D: MEASURING NOT SPECIALLY ADAPTED FOR A SPECIFIC VARIABLE; ARRANGEMENTS FOR MEASURING TWO OR MORE VARIABLES NOT COVERED IN A SINGLE OTHER SUBCLASS; TARIFF METERING APPARATUS; MEASURING OR TESTING NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Aspects related to providing explainable artificial intelligence (XAI) using distributed ledger technology are provided. An XAI platform may train a first iteration of a deep learning model. The platform may store a record of the model to a distributed ledger. The platform may automatically perform an iterative recording process to generate additional iterations of the model and store records of the additional iterations of the model. During the process, the platform may receive outputs of the model. Based on the outputs, the platform may identify an anomalous output. The platform may identify one or more source iterations of the model that correspond to the anomalous output. The platform may initiate one or more response actions based on identifying the one or more source iterations of the model.

## Claims

### Claim 1

A computing platform comprising:
at least one processor;
a communication interface communicatively coupled to the at least one processor; and
memory storing computer-readable instructions that, when executed by the at least one processor, cause the computing platform to:
train a first iteration of a deep learning model based on a training data set;
store, to a distributed ledger, a record of the first iteration of the deep learning model, wherein the record comprises:
first clustering information, wherein the first clustering information identifies clusters of information of the first iteration of the deep learning model; and
first rationale information, wherein the first rationale information identifies digital logic used to generate the clusters of information;
store records of additional iterations of the deep learning model;
identify, based on one or more outputs of the deep learning model, an anomalous output;
identify, by parsing the distributed ledger and based on identifying the anomalous output, one or more source iterations of the deep learning model, wherein the one or more source iterations correspond to the anomalous output; and
initiate, based on identifying the one or more source iterations, one or more response actions configured to prevent additional anomalous outputs.

### Claim 2 (depends on 1)

The computing platform of claim 1, wherein the storing the records of additional iterations of the deep learning model comprises performing an iterative recording process that begins at the first iteration of the deep learning model, as a current iteration, and comprises:
generating a second iteration of the deep learning model;
based on generating the second iteration of the deep learning model, storing a second record of the second iteration to the distributed ledger, wherein the second record comprises second clustering information and second rationale information; and
automatically repeating, in one or more additional iterations of the iterative recording process, the iterative recording process for one or more additional iterations of the deep learning model.

### Claim 3 (depends on 1)

The computing platform of claim 1, wherein the memory stores one or more additional computer-readable instructions that, when executed by the at least one processor, cause the computing platform to:
update, based on initiating the one or more response actions, the deep learning model.

### Claim 4 (depends on 3)

The computing platform of claim 3, wherein the memory stores one or more additional computer-readable instructions that, when executed by the at least one processor, cause the computing platform to:
store, based on updating the deep learning model, a record of the updated deep learning model to the distributed ledger.

### Claim 5 (depends on 3)

The computing platform of claim 3, wherein the memory stores one or more additional computer-readable instructions that, when executed by the at least one processor, cause the computing platform to:
pause, based on identifying the anomalous output, an iterative recording process; and
resume, based on updating the deep learning model, the iterative recording process.

### Claim 6 (depends on 1)

The computing platform of claim 1, wherein the instructions, when executed by the at least one processor, cause the computing platform to identify the anomalous output by:
comparing one or more scores, generated by the deep learning model and corresponding to the one or more outputs, to a threshold score;
identifying, based on the comparing, a score that satisfies the threshold score; and
identifying, as the anomalous output, an output corresponding to the score that satisfies the threshold score.

### Claim 7 (depends on 1)

The computing platform of claim 1, wherein the anomalous output comprises a data point that toggles between different clusters of nodes of the deep learning model.

### Claim 8 (depends on 1)

The computing platform of claim 1, wherein the one or more source iterations each correspond to a unique record, stored in the distributed ledger, comprising an indication that a data point associated with the anomalous output diverged from a first cluster of the deep learning model to a second cluster of the deep learning model.

### Claim 9 (depends on 1)

The computing platform of claim 1, wherein the instructions, when executed by the at least one processor, cause the computing platform to identify the one or more source iterations by:
identifying, by parsing the distributed ledger, a first record indicating that a data point associated with the anomalous output diverged from a first cluster of the deep learning model to a second cluster of the deep learning model; and
selecting, based on identifying the first record, an iteration of the deep learning model corresponding to the first record as the one or more source iterations.

### Claim 10 (depends on 1)

The computing platform of claim 1, wherein the instructions, when executed by the at least one processor, cause the computing platform to identify the one or more source iterations by:
identifying, by parsing each entry of the distributed ledger, a plurality of records indicating that a data point associated with the anomalous output diverged from a first cluster of the deep learning model to a second cluster of the deep learning model; and
selecting, based on identifying the plurality of records, a plurality of iterations of the deep learning model corresponding to the plurality of records as the one or more source iterations.

### Claim 11 (depends on 1)

The computing platform of claim 1, wherein the instructions, when executed by the at least one processor, cause the computing platform to initiate the one or more response actions based on a ruleset identifying one or more parameters for modifying the deep learning model.

### Claim 12 (depends on 1)

The computing platform of claim 1, wherein the one or more response actions comprise one or more of:
removing a feature of the deep learning model,
modifying one or more clusters of neurons of the deep learning model,
reverting to a previous iteration of the deep learning model, or
adjusting one or more weights corresponding to an algorithm of the deep learning model.

### Claim 13

A method comprising:
at a computing device comprising at least one processor, a communication interface, and memory:
training a first iteration of a deep learning model based on a training data set;
storing, to a distributed ledger, a record of the first iteration of the deep learning model, wherein the record comprises:
first clustering information, wherein the first clustering information identifies clusters of information of the first iteration of the deep learning model; and
first rationale information, wherein the first rationale information identifies digital logic used to generate the clusters of information;
storing records of additional iterations of the deep learning model;
identifying, based on one or more outputs of the deep learning model, an anomalous output;
identifying, by parsing the distributed ledger and based on identifying the anomalous output, one or more source iterations of the deep learning model, wherein the one or more source iterations correspond to the anomalous output; and
initiating, based on identifying the one or more source iterations, one or more response actions configured to prevent additional anomalous outputs.

### Claim 14 (depends on 13)

The method of claim 13, wherein the storing the records of additional iterations of the deep learning model comprises performing an iterative recording process that begins at the first iteration of the deep learning model, as a current iteration, and comprises:
generating a second iteration of the deep learning model;
based on generating the second iteration of the deep learning model, storing a second record of the second iteration to the distributed ledger, wherein the second record comprises second clustering information and second rationale information; and
automatically repeating, in one or more additional iterations of the iterative recording process, the iterative recording process for one or more additional iterations of the deep learning model.

### Claim 15 (depends on 13)

The method of claim 13, wherein identifying the anomalous output comprises:
comparing one or more scores, generated by the deep learning model and corresponding to the one or more outputs, to a threshold score;
identifying, based on the comparing, a score that satisfies the threshold score; and
identifying, as the anomalous output, an output corresponding to the score that satisfies the threshold score.

### Claim 16 (depends on 13)

The method of claim 13, wherein identifying the one or more source iterations comprises:
identifying, by parsing the distributed ledger, a first record indicating that a data point associated with the anomalous output diverged from a first cluster of the deep learning model to a second cluster of the deep learning model; and
selecting, based on identifying the first record, an iteration of the deep learning model corresponding to the first record as the one or more source iterations.

### Claim 17

One or more non-transitory computer-readable media storing instructions that, when executed by a computing platform comprising at least one processor, a communication interface, and memory, cause the computing platform to:
train a first iteration of a deep learning model based on a training data set;
store, to a distributed ledger, a record of the first iteration of the deep learning model, wherein the record comprises:
first clustering information, wherein the first clustering information identifies clusters of information of the first iteration of the deep learning model; and
first rationale information, wherein the first rationale information identifies digital logic used to generate the clusters of information;
store records of additional iterations of the deep learning model;
identify, based on one or more outputs of the deep learning model, an anomalous output;
identify, by parsing the distributed ledger and based on identifying the anomalous output, one or more source iterations of the deep learning model, wherein the one or more source iterations correspond to the anomalous output; and
initiate, based on identifying the one or more source iterations, one or more response actions configured to prevent additional anomalous outputs.

### Claim 18 (depends on 17)

The one or more non-transitory computer-readable media of claim 17, wherein the storing the records of additional iterations of the deep learning model comprises performing an iterative recording process that begins at the first iteration of the deep learning model, as a current iteration, and comprises:
generating a second iteration of the deep learning model;
based on generating the second iteration of the deep learning model, storing a second record of the second iteration to the distributed ledger, wherein the second record comprises second clustering information and second rationale information; and
automatically repeating, in one or more additional iterations of the iterative recording process, the iterative recording process for one or more additional iterations of the deep learning model.

### Claim 19 (depends on 17)

The one or more non-transitory computer-readable media of claim 17, storing instructions that, when executed, cause the computing platform to identify the anomalous output by:
comparing one or more scores, generated by the deep learning model and corresponding to the one or more outputs, to a threshold score;
identifying, based on the comparing, a score that satisfies the threshold score; and
identifying, as the anomalous output, an output corresponding to the score that satisfies the threshold score.

### Claim 20 (depends on 17)

The one or more non-transitory computer-readable media of claim 17, storing instructions that, when executed, cause the computing platform to identify the one or more source iterations by:
identifying, by parsing the distributed ledger, a first record indicating that a data point associated with the anomalous output diverged from a first cluster of the deep learning model to a second cluster of the deep learning model; and
selecting, based on identifying the first record, an iteration of the deep learning model corresponding to the first record as the one or more source iterations.